Missing Page Detection Using Learned Image Pairs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image reading apparatuses struggle to detect missing pages in sequences of read images, particularly when page numbers are absent or incorrectly identified, due to 'double feed' issues with auto-document feeders or user errors during manual page flipping.

Innovation Solution

An image reading apparatus employing machine learning to determine missing pages by analyzing pairs of consecutively-generated images using a learned model derived from input data based on image pairs of consecutive and non-consecutive pages, incorporating character and image object data to identify page relationships.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If page number extraction and comparison is used to detect missing pages, then the detection method is simple, but it cannot detect missing pages in originals without page numbers

Engineering Contradiction:
Improvedetection method simplicityVSAvoidapplicability to originals without page numbers
Core Design Contradiction:
Ease of manufactureVSAdaptability or versatility

Solution Approach 1:

The patent replaces the mechanical approach of extracting and comparing page numbers with a machine learning-based image analysis system. The learned model processes image data directly, substituting the need for explicit page number extraction and comparison mechanisms, thereby enabling detection in documents without visible page numbers.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the detection parameters from explicit page number values to learned features from image data. By training the model on image pairs with known consecutiveness labels, the system learns to detect page continuity through visual patterns rather than relying on numerical page indicators.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If machine learning model is used to analyze image pairs for missing page detection, then detection accuracy is improved, but processing complexity increases

Engineering Contradiction:
Improvemissing page detection accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary action by pre-training the machine learning model offline using labeled image pairs. This preliminary training phase creates a ready-to-use learned model that can be deployed without real-time complexity during actual scanning operations, separating the complex learning process from the simpler detection process.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent uses copying by creating a learned model that captures the patterns of page consecutiveness from training data. This model serves as a copy of the knowledge needed to detect missing pages, allowing the system to apply learned patterns without repeatedly performing complex analysis during actual use.

Inventive Principle:
Principle #26Copying

Data Source

PatentUS11206333B2Image reading and learning apparatus, method, and program product for determining a missing page using a learned model and deriving the learned model
Publication Date: 2021.12.21 CANON KK
  • US11206333B2 patent drawing
  • US11206333B2 patent drawing
  • US11206333B2 patent drawing

AI summary

There is provided an image reading apparatus including a reading unit configured to sequentially read respective pages of a target original consisting of a plurality of pages to generate read images, and a determining unit configured to determine whether there is a missing page in the read images. The determining unit is configured to determine whether there is a missing page in the read images by applying, to a learned model, determination data based on a pair of consecutively-generated read images of the target original. The learned model is a model that has been derived through machine teaming using input data based on image pairs of consecutive pages and image pairs of non-consecutive pages of the original for learning, and training data that indicates whether each image pair is a pair of consecutive pages.